Dataset · MCG-NJU
Long video analysis instructions (VideoChat3)
MCG-NJU/VideoChat3-LV116k
An English video-text instruction dataset for training models to understand long videos.
What's inside
An instruction dataset (1,000 to 10,000 examples) created to train the VideoChat3 model with long temporal contexts. Data is obtained using a synthesis pipeline: long videos were filtered by visual quality, semantic content, and temporal coherence, then split into fragments with English text tasks. The dataset complements short clips with tasks where needed details are distributed across different video segments.
What you can do with it
- SFT fine-tuning of multimodal video-text-to-text models for long-video question answering
- Improving model ability to locate scattered events in long videos
- Tasks
- Video text → text
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 22.2K
- Likes
- 15
How to load it
from datasets import load_dataset
ds = load_dataset("MCG-NJU/VideoChat3-LV116k")